通过DataFrames使用spark-csv获取NullPointerException

Question

在spark-csv README中运行的示例Java代码如下: import org.apache.spark.sql.SQLContext；import org.apache.spark.sql.types.*；

SQLContext sqlContext = new SQLContext(sc);
StructType customSchema = new StructType(
    new StructField("year", IntegerType, true), 
    new StructField("make", StringType, true),
    new StructField("model", StringType, true),
    new StructField("comment", StringType, true),
    new StructField("blank", StringType, true));

DataFrame df = sqlContext.read()
    .format("com.databricks.spark.csv")
    .option("inferSchema", "true")
    .option("header", "true")
    .load("cars.csv");

df.select("year", "model").write()
    .format("com.databricks.spark.csv")
    .option("header", "true")
    .save("newcars.csv");

它没有开箱即用地编译，所以经过一番争论，我将不正确的FooType语法改为DataTypes.FooType并将StructFields作为new StructField[]进行了编译；编译器在StructField的构造函数中为metadata请求了第四个参数，但我很难找到关于它含义的文档(javadoc描述了它的用例，但并没有真正决定如何在StructField构造期间传入什么)。使用以下代码，它将一直运行，直到出现任何副作用方法，如collect()

JavaSparkContext sc = new JavaSparkContext(conf);

SQLContext sqlContext = new SQLContext(sc);

// Read features.
System.out.println("Reading features from " + args[0]);
StructType featuresSchema = new StructType(new StructField[] {
    new StructField("case_id", DataTypes.StringType, false, null), 
    new StructField("foo", DataTypes.DoubleType, false, null)
});
DataFrame features = sqlContext.read()
    .format("com.databricks.spark.csv")
    .schema(featuresSchema)
    .load(args[0]);
for (Row r : features.collect()) {
  System.out.println("Row: " + r);
}

我得到了以下异常：

Exception in thread "main" java.lang.NullPointerException
  at org.apache.spark.sql.catalyst.expressions.AttributeReference.hashCode(namedExpressions.scala:202)
  at scala.runtime.ScalaRunTime$.hash(ScalaRunTime.scala:210)
  at scala.collection.immutable.HashSet.elemHashCode(HashSet.scala:65)
  at scala.collection.immutable.HashSet.computeHash(HashSet.scala:74)
  at scala.collection.immutable.HashSet.$plus(HashSet.scala:56)
  at scala.collection.immutable.HashSet.$plus(HashSet.scala:59)
  at scala.collection.immutable.Set$Set4.$plus(Set.scala:127)
  at scala.collection.immutable.Set$Set4.$plus(Set.scala:121)
  at scala.collection.mutable.SetBuilder.$plus$eq(SetBuilder.scala:24)
  at scala.collection.mutable.SetBuilder.$plus$eq(SetBuilder.scala:22)
  at scala.collection.TraversableLike$$anonfun$map$1.apply(TraversableLike.scala:244)
  at scala.collection.TraversableLike$$anonfun$map$1.apply(TraversableLike.scala:244)
  at scala.collection.immutable.HashSet$HashSet1.foreach(HashSet.scala:153)
  at scala.collection.immutable.HashSet$HashTrieSet.foreach(HashSet.scala:306)
  at scala.collection.immutable.HashSet$HashTrieSet.foreach(HashSet.scala:306)
  at scala.collection.TraversableLike$class.map(TraversableLike.scala:244)
  at scala.collection.AbstractSet.scala$collection$SetLike$$super$map(Set.scala:47)
  at scala.collection.SetLike$class.map(SetLike.scala:93)
  at scala.collection.AbstractSet.map(Set.scala:47)
  at org.apache.spark.sql.catalyst.expressions.AttributeSet.foreach(AttributeSet.scala:114)
  at scala.collection.TraversableOnce$class.size(TraversableOnce.scala:105)
  at org.apache.spark.sql.catalyst.expressions.AttributeSet.size(AttributeSet.scala:56)
  at org.apache.spark.sql.execution.datasources.DataSourceStrategy$.pruneFilterProjectRaw(DataSourceStrategy.scala:307)
  at org.apache.spark.sql.execution.datasources.DataSourceStrategy$.pruneFilterProject(DataSourceStrategy.scala:282)
  at org.apache.spark.sql.execution.datasources.DataSourceStrategy$.apply(DataSourceStrategy.scala:56)
  at org.apache.spark.sql.catalyst.planning.QueryPlanner$$anonfun$1.apply(QueryPlanner.scala:58)
  at org.apache.spark.sql.catalyst.planning.QueryPlanner$$anonfun$1.apply(QueryPlanner.scala:58)
  at scala.collection.Iterator$$anon$13.hasNext(Iterator.scala:371)
  at org.apache.spark.sql.catalyst.planning.QueryPlanner.plan(QueryPlanner.scala:59)
  at org.apache.spark.sql.SQLContext$QueryExecution.sparkPlan$lzycompute(SQLContext.scala:926)
  at org.apache.spark.sql.SQLContext$QueryExecution.sparkPlan(SQLContext.scala:924)
  at org.apache.spark.sql.SQLContext$QueryExecution.executedPlan$lzycompute(SQLContext.scala:930)
  at org.apache.spark.sql.SQLContext$QueryExecution.executedPlan(SQLContext.scala:930)
  at org.apache.spark.sql.execution.SQLExecution$.withNewExecutionId(SQLExecution.scala:53)
  at org.apache.spark.sql.DataFrame.withNewExecutionId(DataFrame.scala:1903)
  at org.apache.spark.sql.DataFrame.collect(DataFrame.scala:1384)
...

知道出什么问题了吗？

Answer 1

自述文件似乎非常过时，需要对Java示例进行一些重要的编辑。我跟踪了实际的JIRA which added the metadata field，它指出了Map.empty用例的缺省Java值，编写文档的人一定是直接将Scala翻译成Map.empty，尽管输入参数没有相同的缺省值。

在1.5 branch of SparkSQL's code中，我们可以看到它在没有检查的情况下引用了metadata.hashCode()，这就是导致NullPointerException的原因。Metadata.empty()方法的存在，再加上关于在Scala中使用空映射作为默认映射的讨论，似乎暗示了正确的实现是继续并传递Metadata.empty()，如果您不关心它的话。修改后的示例应该是：

SQLContext sqlContext = new SQLContext(sc);
StructType customSchema = new StructType(new StructField[] {
    new StructField("year", DataTypes.IntegerType, true, Metadata.empty()), 
    new StructField("make", DataTypes.StringType, true, Metadata.empty()),
    new StructField("model", DataTypes.StringType, true, Metadata.empty()),
    new StructField("comment", DataTypes.StringType, true, Metadata.empty()),
    new StructField("blank", DataTypes.StringType, true, Metadata.empty())
});

DataFrame df = sqlContext.read()
    .format("com.databricks.spark.csv")
    .schema(customSchema)
    .option("header", "true")
    .load("cars.csv");

df.select("year", "model").write()
    .format("com.databricks.spark.csv")
    .option("header", "true")
    .save("newcars.csv");

Answer 2

就连我也得到了同样的例外。我通过提供元数据修复了它。

因此，修改代码如下

StructType customSchema = new StructType(
new StructField("year", IntegerType, true,Metadata.empty()), 
new StructField("make", StringType, true,Metadata.empty()),
new StructField("model", StringType, true,Metadata.empty()),
new StructField("comment", StringType, true,Metadata.empty()),
new StructField("blank", StringType, true,Metadata.empty()));

这会解决这个问题